brew/Library/Homebrew/cask/lib/hbc/download_strategy.rb
buo 6a406763f3 Open incomplete download in append mode
Open the incomplete download in append mode instead of write mode.
Opening in write mode truncates the existing file, so curl keeps
restarting downloads instead of resuming the incomplete downloads.
2016-10-25 16:46:59 +09:00

335 lines
9.4 KiB
Ruby

require "cgi"
# We abuse Homebrew's download strategies considerably here.
# * Our downloader instances only invoke the fetch and
# clear_cache methods, ignoring stage
# * Our overridden fetch methods are expected to return
# a value: the successfully downloaded file.
module Hbc
class AbstractDownloadStrategy
attr_reader :cask, :name, :url, :uri_object, :version
def initialize(cask, command = SystemCommand)
@cask = cask
@command = command
# TODO: this excess of attributes is a function of integrating
# with Homebrew's classes. Later we should be able to remove
# these in favor of @cask
@name = cask.token
@url = cask.url.to_s
@uri_object = cask.url
@version = cask.version
end
# All download strategies are expected to implement these methods
def fetch; end
def cached_location; end
def clear_cache; end
end
class HbVCSDownloadStrategy < AbstractDownloadStrategy
REF_TYPES = [:branch, :revision, :revisions, :tag].freeze
def initialize(cask, command = SystemCommand)
super
@ref_type, @ref = extract_ref
@clone = Hbc.cache.join(cache_filename)
end
def extract_ref
key = REF_TYPES.find do |type|
uri_object.respond_to?(type) && uri_object.send(type)
end
[key, key ? uri_object.send(key) : nil]
end
def cache_filename
"#{name}--#{cache_tag}"
end
def cache_tag
"__UNKNOWN__"
end
def cached_location
@clone
end
def clear_cache
cached_location.rmtree if cached_location.exist?
end
end
class CurlDownloadStrategy < AbstractDownloadStrategy
# TODO: should be part of url object
def mirrors
@mirrors ||= []
end
def tarball_path
@tarball_path ||= Hbc.cache.join("#{name}--#{version}#{ext}")
end
def temporary_path
@temporary_path ||= tarball_path.sub(/$/, ".incomplete")
end
def cached_location
tarball_path
end
def clear_cache
[cached_location, temporary_path].each do |f|
next unless f.exist?
raise CurlDownloadStrategyError, "#{f} is in use by another process" if Utils.file_locked?(f)
f.unlink
end
end
def downloaded_size
temporary_path.size? || 0
end
def _fetch
odebug "Calling curl with args #{cask_curl_args.utf8_inspect}"
curl(*cask_curl_args)
end
def fetch
ohai "Downloading #{@url}"
if tarball_path.exist?
puts "Already downloaded: #{tarball_path}"
else
had_incomplete_download = temporary_path.exist?
begin
File.open(temporary_path, "a+") do |f|
f.flock(File::LOCK_EX)
_fetch
f.flock(File::LOCK_UN)
end
rescue ErrorDuringExecution
# 33 == range not supported
# try wiping the incomplete download and retrying once
if $CHILD_STATUS.exitstatus == 33 && had_incomplete_download
ohai "Trying a full download"
temporary_path.unlink
had_incomplete_download = false
retry
end
msg = @url
msg.concat("\nThe incomplete download is cached at #{temporary_path}") if temporary_path.exist?
raise CurlDownloadStrategyError, msg
end
ignore_interrupts { temporary_path.rename(tarball_path) }
end
tarball_path
rescue CurlDownloadStrategyError
raise if mirrors.empty?
puts "Trying a mirror..."
@url = mirrors.shift
retry
end
private
def cask_curl_args
default_curl_args.tap do |args|
args.concat(user_agent_args)
args.concat(cookies_args)
args.concat(referer_args)
end
end
def default_curl_args
[url, "-C", downloaded_size, "-o", temporary_path]
end
def user_agent_args
if uri_object.user_agent
["-A", uri_object.user_agent]
else
[]
end
end
def cookies_args
if uri_object.cookies
[
"-b",
# sort_by is for predictability between Ruby versions
uri_object
.cookies
.sort_by(&:to_s)
.map { |key, value| "#{CGI.escape(key.to_s)}=#{CGI.escape(value.to_s)}" }
.join(";"),
]
else
[]
end
end
def referer_args
if uri_object.referer
["-e", uri_object.referer]
else
[]
end
end
def ext
Pathname.new(@url).extname
end
end
class CurlPostDownloadStrategy < CurlDownloadStrategy
def cask_curl_args
super
default_curl_args.concat(post_args)
end
def post_args
if uri_object.data
# sort_by is for predictability between Ruby versions
uri_object
.data
.sort_by(&:to_s)
.map { |key, value| ["-d", "#{CGI.escape(key.to_s)}=#{CGI.escape(value.to_s)}"] }
.flatten
else
["-X", "POST"]
end
end
end
class SubversionDownloadStrategy < HbVCSDownloadStrategy
def cache_tag
# TODO: pass versions as symbols, support :head here
version == "head" ? "svn-HEAD" : "svn"
end
def repo_valid?
@clone.join(".svn").directory?
end
def repo_url
`svn info '#{@clone}' 2>/dev/null`.strip[/^URL: (.+)$/, 1]
end
# super does not provide checks for already-existing downloads
def fetch
if tarball_path.exist?
puts "Already downloaded: #{tarball_path}"
else
@url = @url.sub(/^svn\+/, "") if @url =~ %r{^svn\+http://}
ohai "Checking out #{@url}"
clear_cache unless @url.chomp("/") == repo_url || quiet_system("svn", "switch", @url, @clone)
if @clone.exist? && !repo_valid?
puts "Removing invalid SVN repo from cache"
clear_cache
end
case @ref_type
when :revision
fetch_repo @clone, @url, @ref
when :revisions
# nil is OK for main_revision, as fetch_repo will then get latest
main_revision = @ref[:trunk]
fetch_repo @clone, @url, main_revision, true
fetch_externals do |external_name, external_url|
fetch_repo @clone + external_name, external_url, @ref[external_name], true
end
else
fetch_repo @clone, @url
end
compress
end
tarball_path
end
# This primary reason for redefining this method is the trust_cert
# option, controllable from the Cask definition. We also force
# consistent timestamps. The rest of this method is similar to
# Homebrew's, but translated to local idiom.
def fetch_repo(target, url, revision = uri_object.revision, ignore_externals = false)
# Use "svn up" when the repository already exists locally.
# This saves on bandwidth and will have a similar effect to verifying the
# cache as it will make any changes to get the right revision.
svncommand = target.directory? ? "up" : "checkout"
args = [svncommand]
# SVN shipped with XCode 3.1.4 can't force a checkout.
args << "--force" unless MacOS.version == :leopard
# make timestamps consistent for checksumming
args.concat(%w[--config-option config:miscellany:use-commit-times=yes])
if uri_object.trust_cert
args << "--trust-server-cert"
args << "--non-interactive"
end
args << url unless target.directory?
args << target
args << "-r" << revision if revision
args << "--ignore-externals" if ignore_externals
@command.run!("/usr/bin/svn",
args: args,
print_stderr: false)
end
def tarball_path
@tarball_path ||= cached_location.dirname.join(cached_location.basename.to_s + "-#{@cask.version}.tar")
end
def shell_quote(str)
# Oh god escaping shell args.
# See http://notetoself.vrensk.com/2008/08/escaping-single-quotes-in-ruby-harder-than-expected/
str.gsub(/\\|'/) { |c| "\\#{c}" }
end
def fetch_externals
`svn propget svn:externals '#{shell_quote(@url)}'`.chomp.each_line do |line|
name, url = line.split(/\s+/)
yield name, url
end
end
private
# TODO/UPDATE: the tar approach explained below is fragile
# against challenges such as case-sensitive filesystems,
# and must be re-implemented.
#
# Seems nutty: we "download" the contents into a tape archive.
# Why?
# * A single file is tractable to the rest of the Cask toolchain,
# * An alternative would be to create a Directory container type.
# However, some type of file-serialization trick would still be
# needed in order to enable calculating a single checksum over
# a directory. So, in that alternative implementation, the
# special cases would propagate outside this class, including
# the use of tar or equivalent.
# * SubversionDownloadStrategy.cached_location is not versioned
# * tarball_path provides a needed return value for our overridden
# fetch method.
# * We can also take this private opportunity to strip files from
# the download which are protocol-specific.
def compress
Dir.chdir(cached_location) do
@command.run!("/usr/bin/tar",
args: ['-s/^\.//', "--exclude", ".svn", "-cf", Pathname.new(tarball_path), "--", "."],
print_stderr: false)
end
clear_cache
end
end
end